So far most reviews have focused on performance - and I'll be messing about with that later - but thought I'd focus a bit on other bits and pieces. For those who are relatively new to this thread, please note that this is for basically an engineering sample and is not necessarily the final product.
Updates for 2015/07/08 with feedback from designers
Beauty shots as typically used in my StickMiners thread.
Documentation
There is none - well, this and the development thread. Hopefully something will be set up at http://gekkoscience.com/ in due time
On the other hand: if you've ever used a different small Bitcoin miner, then documentation is practically unnecessary.
Plugging it in
As a USB device, the Compac can plug straight into a USB port, but due to its width and power requirements, a suitable USB hub is recommended.
Software
Using cgminer, the Compac is recognized as an Antminer U3 after the default drivers - Silicon Labs CP2102 drivers - are replaced using Zadig. Only a single device configuation parameter is needed:
Code:
cgminer -o stratum+tcp://pool_host:pool_port -u username -p password --au3-freq FREQUENCY
The frequencies accepted via cgminer by default are 100 through 250 in these steps:Code:
100, 125, 150, 175, 193.75, 196.88, 200, 206.25, 212.5, 218.75, 225, 237.5, 243.75, 250
Further frequency support can be compiled in, and is being worked on by GekkoScience. Higher frequencies do tend require a higher voltage, which is set using a piece of hardware on the board; the cgminer --au3-volt parameter is effectively ignored. Higher frequencies also draw more power.Using bfgminer - well, unfortunately I got no further than a single blip of the board's LED and bfgminer throwing an error, no matter the combination of scan settings and drivers:
Code:
Failed to sanity check in lowl-vcom.c windows_usb_get_string():508
bfgminer support can probably be added in the future, but cgminer+zadig works quite well.
Tweaking
The Compac's performance is effectively regulated by three things: its operating frequency, the core voltage, and its temperature.
A higher frequency demands a higher core voltage, uses more power, which makes the chip hotter which may cause increased hardware errors (or worse).
A lower frequency can do with a lower core coltage, uses less power, which lets the chip run cooler.
However, too low a core voltage and the miner will not initialize or cause increased hardware errors.
See the Core voltage adjustment section for a bit more detail.
Dimensions
How big is the Compac? Overall dimensions are approximately 93mm x 25mm x 14.5mm, and it weighs about 25g.
How does that compare to some other miners? Bigger than a Block Erupter USB, AntMiner U1/U2 and similar, thinner but longer than a NanoFury 6, but much smaller than a Twin Chip Fury.
( The black bit sticking out the side of the GekkoScience Compac is a female header I soldered on. )
Thickness-wise, it is only marginally thicker than an Antminer U2.
Weight-wise, it is about as heavy as an Antminer U2, a bit less than double that of a Block Erupter USB (14.5g), and a bit less than half that of a Twin Chip Fury (59g).
Heat sink and thermals
The heat sink is essentially the same size as that of the Antminer U2 heat sink, but has 8 fins instead of 7.
The heat sink is bolted down with 3 screws which are pretty easy to remove. The heat sink sits on top of the Bitmain Technologies BM1384 chip as it's a flip chip design, meaning most of the heat is best evacuated out the top. The heat sink does overlap the board, something necessitated due to the tall components elsewhere on the board. An extensive redesign might alleviate that, but from the development thread, earlier layouts had issues of their own. In the end, this is primarily an aesthetic aspect.
The heat sink also sits a bit off the board as it has to connect to the BM1384 chip. A few surrounding components (capacitors) are actually a tiny bit taller than the chip, so the heat sink has a small metal spacer to make sure that the heat sink has good thermal contact, instead of sitting on top of those capacitors and 1. not contacting the chip and, worse, 2. potentially shorting out those capacitors.
The heat sink is not the final color: it is planned to be green, at least from GekkoScience themselves, so it might look a little more like this:
GekkoSciencce: I think the actual green for the heatsinks will be more "evergreen"
Third party sellers - e.g. the German licensor - may opt for a different color, and the sides of the heat sink provide ample branding space, be it a simple Dymo labelmaker sticker, silkscreen printing or (laser) engraving.
Although the BM1384 is a flip chip, the bottom does get warm as well. Here's a video of a failed initialization: https://www.youtube.com/watch?v=mWeMsrdrR-8
Note that the thermal tape used here has a range of about 25°C - 35°C, so by no means should this be interpreted as it actually getting hot. Nevertheless, for those wishing to clock higher, an additional heat sink or heat spreader on the bottom may not be a terrible idea, as long no short circuits are created.
The top of the board also stays fairly cool, although the CP2102 USB serial chip and one of the voltage regulators do get relatively warm even without any load. When under load, the other components warm up a bit as well, but nowhere near as much as was apparent for the Block Erupter USB test.
LEDs
When plugged in, the Compac will light the LED bright green. This LED also stays on when the board is hashing away - unlike the Block Erupter USB and a few other miners. I have to admit that it is sufficiently bright that I put thermal tape on the LED as well to make it a bit less so. I would recommend a larger resistor value for the green LED to make it a bit more dim and save (a small) amount on current used. Of course end-users can also take a sharpie to the LED or cover it up.
When the board is hashing away and finds a share, the LED will appear to blink in a bluish white, courtesy of what at first glance may look like a purple color.
This is actually a combination of both blue and orange/amber (not quite red) LED chips within the package.
The blue and orange/amber channels appear to be soldered together, so if you were wondering if there were further statuses - this does not appear to be the case.
Core voltage adjustment
The core voltage can be adjusted using the trim potentiometer - in the corner of the board opposite that of the GekkoScience logo - using a small screwdriver.
Clockwise = a lower resistance = lower core voltage.
GekkoScience: [A] thing changed on the V0.5 is the pot rotation direction; it'll be back to clockwise increases voltage instead of counterclockwise
If the core voltage is too low, the board will fail to initialize - after it has initialized, however, the core voltage can be brought down. This makes testing a bit more difficult than set-and-forget if trying to eke out the lowest power use for a given hash rate.
So far I've found that a 2 o'clock position always facilitates initialization, while 3 o'clock is a bit more prone to failure. This is likely to vary as trim potentiometers tend to have a fair bit of tolerance and thus differences between components out of the factory.
One thing to note is that the trimpot of choice does not have detents. What this means is that when you turn it clockwise too far, you can enter a 'dead zone' on the resistance track, in effect bringing the resistance up to the maximum and with it the core voltage. If you are not using a USB meter or test equipment to monitor the power draw or core voltage directly, you may thus be led to believe you're lowering the core voltage when in reality it just shot up. Not an issue for tinkerers, but still something to keep in mind.
Core voltage / testing
The core voltage can be monitored directly using the exposed pads on the bottom of the board - although I would recommend soldering a header, preferably female, to those ports to make it easier to hook up some test leads. Using the core voltage is, arguably, a better measure for exact testing than trim potentiometer angles or power drawn, as the angles are not quite exact and power draw may fluctuate.
There are 5 pads in total , although I have to admit I haven't quite deciphered one of them as of yet; it seems to short to ground.
GekkoScience: The question-mark pad is indeed GND [...] The final version for production has a single GND pad and RESET is broken out on the fifth pad in case someone needs it.
These pads are, as mentioned, exposed and sit close to any surface the Compac is rested on. If you were to use multiple Compacs in a fairly dense USB hub, you may wish to enure that the heat sink of one can never short out the pads of another, or put some isolating material between them. For the production version, I would suggest not exposing these pads by default, leaving off any ENIG/gold plating and keeping solder mask on top of them. This does mean tinkerers will have to expose the pads specifically if they want to, but makes it a bit less prone to accidental shorts.
Engineering / build quality
I'm going to re-stress that this is effectively an engineering sample, so some of the feedback here is likely to not even apply to the production version. I'm also going to be semi-harsh, because why not.
The heat sink is reasonable quality. The screw taps go through the entire thickness which is a bit overkill, and burrs are left around the taps. This shouldn't pose any practical problems as long as very loose burrs are brushed away to prevent any of them falling out and accidentally shorting things.
GekkoScience: They were samples sent from the factory, drilled from stock they already had run out.
The components are all good quality. No cracked inductors here.
Component and trace layout is decent. I can't really comment on this as I know GekkoScience faced issues earlier on and some of the layout decisions may have stemmed from alleviating those issues. There's a few 90° angles that could probably be avoided, but with modern fabrication and them not being part of any high frequency signalling paths, not really an issue.
The PCB itself is of pretty good quality. It's a 4-layer 1.6mm board, with good alignment of solder mask, copper and drills.
Soldering is decent, given that it was done outside of a professional setup. There are quite a few solder balls around, but none of them posing an acute problem. GekkoScience have already mentioned that they're getting a nice SMD soldering station, so this should not be an issue going into production.
GekkoScience: We have inbound [...] a pick-and-place and an IR oven so placement and soldering should be a fair bit more precise on the end product.
The silkscreen printing is minimalistic - i.e. there's no component indicators let alone values. I understand that future versions have the pads on the bottom labeled, at least.
GekkoScience: On the final version, the pads on the back are labeled.
Not really applicable to the Compac itself, but: packaging for shipment was also quite good. Antistatic bag, bubble wrap, wedged diagonally in a sturdy box so it wasn't going to shift around despite typical postal treatment.
Sub-verdict for this post
The GekkoScience Compac is a solid StickMiner, certainly has raised a few eyebrows in regard to the performance, and in terms of tinkering allows a fair bit of control for a trade-off of speed, power usage, and acceptable HW error rates. A few niggles aside - that may or may not be addressed in production version(s) - this is a miner that most users should be able to use almost entirely out of the box.
One might even say that the GekkoScience Compac is the miner that the Antminer U3 should have been.
With the Antminer U3 instead being the Antminer P1, and preferably tested better
